Item: Anthropic
Author: Karl Strauchman

Anthropic shipped Claude Opus 5 on July 24, holding Opus 4.8's pricing of $5 per million input tokens and $25 per million output tokens while claiming it more than doubles its predecessor's Frontier-Bench v0.1 score and lands within 0.5% of Fable 5 on CursorBench 3.2 at max effort. It's the default on Claude Max and the top-tier option on Claude Pro. Context is 1M tokens; max output is 128K.

The pricing is the story. Opus 5 doesn't undercut Fable 5 by trimming a margin, it undercuts it by roughly half on CursorBench 3.2 while trailing by less than a percentage point at max effort. On ARC-AGI 3, Anthropic says the model scores three times the next-best result. On Zapier's AutomationBench it topped the leaderboard without spending more tokens than prior Claude models and hit 100% on a churn-prevention sequence previous models failed outright. On a FreeCAD-based computer-vision task, no competing model solved it in five attempts.

Anthropic's framing is unusually candid about scope. "Opus 5 as your daily driver, the model you hand complex work to and review when it's done. Fable 5 for your most ambitious work, the days-long autonomous projects nothing could take on before," an Anthropic spokesperson told VentureBeat. And, on the benchmarks themselves: "The evals where Opus 5 wins are bounded tasks with a specific outcome... What those evals don't measure is duration."

That's a rare admission in a launch cycle usually built around leaderboard maximalism. It says the bounded-task frontier and the long-horizon frontier are diverging, and Anthropic is content to sell the cheaper half at scale while Fable 5, and the still-restricted Mythos 5, sit at the top of the price sheet. Mythos 5 remains available only to a select group of organizations after the Trump administration briefly pulled it and Fable 5 in June over export-control concerns.

The commercial logic Bloomberg draws out is enterprise cost discipline meeting Chinese competition, most recently Moonshot's open-weight Kimi K3. The same dynamic that drove AWS's price cuts through the 2010s is now operating on inference: when a good-enough substitute exists at zero marginal license cost, the frontier vendor sells efficiency, not ceiling.

The plumbing changes are consistent with that read. Mid-conversation tool changes are in beta across Fable 5, Mythos 5, Opus 4.8, and Opus 5 via a dated header. Fast mode has been removed from Opus 4.7 entirely; requests with speed: "fast" now error rather than fall back, and operators are told to migrate to Opus 5 or stay on Opus 4.8. On the CNBC-noted retention point, Opus 5 doesn't retain user data for 30 days the way Fable 5 does, a detail that'll matter to procurement before any benchmark does.

The frontier hasn't moved. The floor under it has.

Sources

https://www.anthropic.com/news/claude-opus-5
https://www.bloomberg.com/news/articles/2026-07-24/anthropic-unveils-more-cost-efficient-model-for-everyday-tasks
https://www.cnbc.com/2026/07/24/anthropic-claude-opus-5-ai-fable-5-cost.html
https://venturebeat.com/orchestration/anthropic-launches-claude-opus-5-a-cheaper-ai-model-for-coding-agents-and-enterprise-workflows
https://support.claude.com/en/articles/12138966-release-notes